AI capabilities

Alloy separates predictive fraud intelligence from an agentic assistant for procedural work. The assistant uses the context of its decisioning and identity platform.

• Resolve review tasks
• Research business entities
• Explain pending applications

Mesh documents agents that resolve routine screening alerts under configured thresholds and risk policies. The published scope includes sanctions, PEP, and adverse-media review, with recorded decision reasons.

• Investigate screening matches
• Resolve eligible routine alerts
• Record evidence and decision reasons

CorShield and payment models produce risk assessments. AI also classifies merchant activity, analyzes site imagery and content, and summarizes review evidence. Assistants support an analyst; separately configured agents can resolve routine alerts, request information, and trigger connected payout controls. Coris MCP exposes both read and write workflows to a connected AI client. Public pages describe these actions, but protected technical docs prevent verification of the full permission and execution contract.

• Assess business impersonation and payment risk
• Classify business activity and inspect website or location evidence
• Summarize cases, resolve permitted alerts, and escalate exceptions

Integrity Check combines Dow Jones risk content with Xapien generative AI, language processing, and entity resolution. It prepares research reports. This does not establish a native autonomous AML case or filing agent.

• Research people and companies
• Resolve relevant source records
• Summarize due diligence findings

Feedzai combines transaction and behavioral models with RiskOps Studio assistance for documentation, strategies, and investigations. Public guidance features do not establish autonomous case clearance or policy publication.

• Score transaction risk
• Prioritize AML alerts
• Explain documentation and investigation context

FinScan emphasizes data preparation and field-level matching rules. Advanced matching is not evidence of a predictive fraud model or a generative investigation agent.

• Identify hidden party names
• Improve candidate matching
• Reduce duplicate review work

AI Forensics supports transaction and screening review. Published controls include modes from silent evaluation to more automated handling, with explanation and version information.

• Draft rule logic
• Investigate monitoring alerts
• Prepare screening recommendations

Percy supports evidence gathering and case analysis. A playbook can run a released agent version and pass its structured output to rules and action nodes. Those nodes can set onboarding decisions, request review, or update status, tags, and notes. This documents bounded workflow automation, not unrestricted authority to clear compliance cases, move funds, or file reports.

• Research a business or screening hit
• Return structured findings to a configured playbook
• Support cited investigations and historical testing

Hawk's AML Investigative Agent assembles evidence, updates summaries, and recommends or performs permitted routine steps. Customer Screening Agent separately evaluates hits under the institution's policy.

• Assemble AML investigation evidence
• Review screening hits
• Recommend or run permitted investigation steps

Jumio describes AI for document verification, biometrics, identity connections, and screening. These are specific verification functions, not proof of autonomous compliance investigation.

• Detect altered documents
• Assess liveness and selfie matching
• Link identity risk signals

ThreatMetrix applies AI models to digital fraud patterns. Screening products use their own matching and decision tools. The AI method and operational authority differ by product.

• Score digital identity risk
• Prioritize screening matches
• Support automated risk decisions

LSEG combines researched risk data with screening technology. Do not equate a match score with a probability that a person is sanctioned.

• Retrieve candidate matches
• Support identity disambiguation
• Prioritize screening review

The Due Diligence Agent coordinates research and report drafting using Moody’s data and selected external sources. It produces structured findings with source attribution. Analysts own the final conclusions.

• Retrieve due diligence evidence
• Prepare sourced investigation reports
• Support customer risk review

ActOne InvestigateAI prepares investigation plans and summaries. NarrateAI drafts SAR narratives. Xceed has separate agent capabilities; permissions and product access must be checked for each deployment.

• Plan and summarize investigations
• Draft SAR narratives
• Support product-specific investigation steps

The recommended logic-v2 matcher uses deterministic rules, culturally aware name matching, and identifier checks. It returns candidates and similarity scores; it is not an AML investigator or a predictive fraud model.

• Find candidate entities
• Score attribute similarity
• Support source-level investigation

Compliance Studio supplies ML and graph analytics. Compliance Agent tests monitoring controls in a simulation. AI Investigator separately assists live investigation work. Each function has a different purpose and approval boundary.

• Find connected activity
• Test control blind spots
• Assist investigation review

Oscilar describes agents for building policies, analyzing performance, and investigating cases, alongside ML models used in risk decisions.

• Build decision workflows
• Analyze policy performance
• Prepare case investigations

Amigo Insights supplies analytics. Vendor enrichment has documented automatic and review modes; AI outreach is opt-in beta. The product site also describes risk models. Amigo Risk, Chargebacks, and Onboarding remain labeled Coming Soon. These are different evidence levels and operating permissions.

• Answer payment-data questions and prepare charts
• Research vendor payment methods and fill empty record fields when configured
• Schedule enabled beta outreach to collect payment preferences

Policy Agents extracts requirements, flags conflicts, and maps them to Persona workflows and templates. Persona Assistant can propose and apply configuration changes after human approval. This is a specific policy task, distinct from automatic case clearance.

• Map policy requirements
• Find configuration gaps
• Propose approved configuration changes

The fraud guide describes configurable strike-based scores and relationships between user and device identifiers. User insights infer recurring locations, with explicit uncertainty. Neither establishes a predictive payment-fraud model, a generative investigator, or an autonomous AI agent. Treat those capabilities as unconfirmed until Radar provides product-specific evidence.

• Return device, network, and location risk indicators
• Aggregate configured user/device risk history
• Supply evidence for an application-owned step-up or access decision

Sardine lists agents for rule help, data analysis, due diligence, screening, transaction review, and SAR generation. Each has a different evidence and approval requirement.

• Research businesses
• Investigate fraud networks
• Prepare screening and SAR work

The reviewed product material establishes anomaly detection, ML, explainability, and network analysis. It does not establish a specific autonomous AML agent included in every deployment.

• Detect unusual behavior
• Find hidden relationships
• Explain analytical flags

SEON combines fraud models with explained screening-hit recommendations and AI-generated regulatory-report drafts. FinCEN transmission is a configured reporting workflow, not evidence that an AI agent may file without review.

• Score fraud using configured signals
• Explain screening-hit recommendations
• Draft report narratives from case evidence

Sift applies AI/ML to network and customer events. ActivityIQ in Account Defense provides generative summaries of cross-session behavior. This is review assistance; independent case or filing authority is not established.

• Score payment fraud
• Detect account takeover
• Summarize suspicious sessions

Predictive models assess identity and document risk. Separately, the RiskOS AI Suite provides business research, case recommendations, rule assistance, and explanations. Socure Agent Intelligence identifies AI actors; it is a different product.

• Research businesses
• Recommend case outcomes
• Explain scores and policy changes

Radar uses predictive models for transaction, merchant, and abuse risk. Radar Assistant uses an LLM to draft transaction rules. Connected-account agentic insights explain risk for investigation, while the preview website signal uses LLM analysis. Model-driven blocking and scheduled reserve actions are operational automation; the reviewed sources do not establish that a generative agent can independently publish policy, reject merchants, or move funds.

• Score payments and identify changes in fraud pressure
• Draft rules for user testing and activation
• Explain connected-account risks and preview website findings

Sumsub combines verification and transaction models with graph-based fraud detection. Summy separately summarizes cases and helps prepare workflow and rule settings. Human reviewers retain final judgment.

• Analyze verification, transaction, and connected-fraud context
• Summarize applicant and case evidence
• Draft reviewed rules and verification-flow settings

RAY gathers alert and customer data, compares behavior with KYC profiles, and prepares an investigation report. Its assistant summarizes documents and explains case context. Analysts retain control of final decisions.

• Assemble case evidence
• Prepare investigation reports
• Explain alert context

Trulioo combines identity checks with UBO Discovery, which uses governed sources and LLM-assisted relationship mapping to reconstruct ownership. Unresolved evidence remains a gap for review; the reviewed scope does not establish independent compliance clearance.

• Resolve business entities and ownership relationships
• Reconstruct source-linked ownership chains
• Expose unresolved or conflicting evidence for review

Unit21 combines evidence collection and narrative drafts with rule generation and backtested tuning proposals. MCP separately provides permitted read access. Analysts approve filings and rule publication.

• Alert and case evidence review
• SAR narrative and filing-package preparation
• Rule Writer drafts
• Insights and Optimization recommendations

Verafin describes sanctions and EDD review workers, plus AML cash-structuring and ACH fraud skills. Optional disposition depends on workflow and enablement. The public rollout plan does not establish universal availability.

• Review sanctions hits
• Research high-risk customers
• Triage supported AML and fraud alerts

Veriff uses AI and fraud analysis across documents, biometrics, devices, and session context. Screening adds separate data and matching functions.

• Find document manipulation
• Detect face and session anomalies
• Link suspicious sessions